2-D Convolution, Decaying Attention, and Training Loop

Requirements 2-D Convolution in NumPy Using the following arrays, implement a valid 2-D convolution with stride 1: The result must have shape (H - kH + 1, W - kW + 1). Begin with an implementation based on nested loops, and then discuss how slicing or vectorization could be used. Decaying Attention Implement the following attention computation: The bias matrix b should depend on the absolute distance between token positions. State whether k has already been transposed; under…